Understanding Your Results

NLPatent is an AI-based semantic search platform, which is vastly different from traditional keyword search databases. So, for those who have years of experience on legacy platforms, understanding your results in NLPatent can take some getting used to. Whether you are a boolean search expert, or just starting out, this article serves to help better understand how results are displayed in NLPatent.

How are Results Ranked?

When a search query is submitted, our AI ranks all patents in the database from most relevant to least relevant, based on conceptual relevance to the query. By default, we display the top 100 results, as this is typically sufficient for most search types. If you’d like to see more results in the NLPatent platform, you can change the maximum number of results in the bottom right corner of the search tab.

On the results page for either a Natural Language or Patent Number search, the position of each patent (e.g., 1, 2, 3) indicates its conceptual relevance to the query as determined by the AI. For instance, the patent in position #1 is the most relevant, #2 is the second most relevant, and so on.

This ranking system differs from keyword searches, where patents are ranked based on the frequency of keyword mentions. However, a higher keyword count does not necessarily equate to higher relevance.

How Filters Change the Results Obtained

After starting a search, you'll see the number of searched patents displayed in the top right corner, such as "X M searched," representing the total universe of patents being considered.

Without Filters:

If no filters are applied, the AI will search the entire database, ranking all patent documents from most to least relevant. As of the date this article was written, this would mean 123 million patents would be searched, displaying the top 100 results in the platform’s interface.

With Filters:

When filters are applied, the number of searched patents decreases to reflect the filtered subset of patents that match the specified criteria. For example, applying jurisdiction filters for WIPO and USA will reduce the number to 24 million, representing documents that are either part of WIPO or the USPTO. The AI will then rank these 24 million publications and provide the top 100 most relevant documents.

Similarly, applying a keyword filter will further narrow the results. For instance, using "Device" in “All Fields” as a keyword filter will restrict the search to patents that mention the word "Device." The number of searched patents will decrease again, showing only those that are part of WIPO, USPTO, and include the keyword "Device" at least once anywhere in the specification.

In summary, filters restrict the universe of patent documents the AI considers in determining relevance, resulting in a decreased number of searched patents according to the specified criteria.

Understanding your Results

While the AI displays the most relevant results, it can sometimes be challenging to understand why certain documents are identified. This is where our Relevance Analysis feature comes into play. This tool helps users understand why a document is considered relevant by listing specific similarities and differences and pinpointing pertinent points of relevance within the document, which may include passages, claim numbers and quotes from the specification.

Relevance Analysis compares the query to the identified relevant results, highlighting which features in the query are present and which are different. This feature works best when more detailed information is provided at the query level, which is the case for a Natural Language search. It is important to note, however, that Relevance Analysis is independent of any refinements and always compares the query directly to the patent document.

Putting it together

In summary, NLPatent displays the most conceptually relevant patents at the top of the list, making it an excellent tool to identify the closest references to any given query or patent document very quickly and efficiently. If you are having difficulty finding relevant references, consider the following:

The search query might need improvement. [See our best practices articles: Natural Language Search].
Filters and keyword restrictions may be excluding otherwise relevant results.
You might be operating in a whitespace technology area.